A Simple and Reliable Method for Estimating Building-Scale Height Based on Multisource Datasets

Abstract

Building height dataset is crucial in urban studies, holding significant importance in understanding the interactions of human activities and the built-up environment. However, high-resolution three-dimensional building datasets covering large areas are limited. A rapid and accurate method for revealing fine-scale urban morphology is required. In this study, we developed a method for estimating building heights at the building scale. First, we integrated multisource datasets (i.e., synthetic aperture radar, optical, terrain, social-economic, and vector-based datasets) and built the machine learning model for building height estimation. Second, we applied the model to 11 cities in the U.S. and assessed its performance. Our results were consistent with the reference data, indicating that the effectiveness of our method is applicable [i.e., the R2 was 0.82, and the root mean square error (RMSE) was 3.39 m]. The evaluated results in various cities, across different height intervals, and within distinct regions also show the good agreement with reference heights according to the correlation (R2: 0.51–0.86, RMSE: 2.57–5.97 m in cities) and similar height distribution. Moreover, our results also showed the superiority by comparing with other height datasets at different scales. Finally, we mapped the building-scale height to characterize the urban morphology. These results demonstrate our proposed method's usable accuracy and the vast application potential in estimating building heights. Our proposed method's refined building height information can significantly help socioeconomic and climatological urban studies.

Keywords

• Building height
• high-resolution
• multisource datasets
• urban morphology
• urbanization